Development of a fault-tolerant model predictive controller for vehicle lateral stability

Abstract

Recently, there has been an increased interest in the automotive industry using scaled test vehicles to test the real-time performance of modeling and control algorithms. A scaled prototype, specifically developed and instrumented for enhancing vehicle lateral stability, offers distinct advantages in terms of cost reduction and the ability to repeat tests under various vehicle maneuvering scenarios rapidly. First, the mechatronic design of a 1:8 scaled electric vehicle with 4-wheel-drive and 4-wheel-independent-steering was done and the prototype vehicle was built. Plant model parameters such as the cornering coefficients of the tires are estimated using various methods such as traditional neural network training, a Physics Informed Deep Learning (PIDL) algorithm, and Pacejka’s tire modeling procedure. Secondly, a fault-tolerant reconfigurable model predictive controller (MPC) is proposed to enhance reference tracking for four-wheel-drive and four-wheel-steering vehicles under concurrent steering actuator faults. The method detects, isolates, and estimates fault magnitudes, which inform adjustments to the MPC formula. Performance validation is conducted through obstacle avoidance maneuvers with a control-oriented vehicle model and real-time applicability tests with a Processor-in-the-Loop system using a high-fidelity vehicle model. The test results confirm the proposed algorithm’s superior performance over the conventional MPC. Lastly, a computationally efficient two-path optimal control allocation method is proposed to reduce controller block execution time in vehicle ECU. High-fidelity results prove the computational cost reduction of the proposed algorithm over the conventional allocation method.